Anbn. Behav., 1992, 43, 585-593
Memory for cache locations in Merriam's kangaroo rats
L U C I A F. J A C O B S *
Department of Psychology, University of Toronto, Toronto, ON M5S 1A1 Canada
(Received9 January 1990; initialacceptance5 June 1991;
final acceptance 10 September 1991; MS. number: A5715)
Abstract. The ability of Merriam's kangaroo rats, Dipodono's merriami, to remember the location of food
caches and to relocate caches in the absence of the odour of buried seeds was examined. Eight wild-caught
kangaroo rats cached seeds in an experimental arena, and retrieved them 24 h later. Before retrieval, all
odours associated with the cache sites were removed and seeds were replaced in only halfofthe cache sites.
During retrieval, kangaroo rats were significantly more likely to search cache sites, with or without seeds,
than non-cache sites. Non-cache sites were primarily investigated after all cache sites had been searched,
indicating that search of non-cache sites did not denote an error in cache retrieval. These results suggest
that kangaroo rats can remember the locations of food caches, and can relocate cache sites even when there
is no odour of buried seeds. To estimate the advantage enjoyed by the forager with greater information, a
second experiment compared an owner's success in retrieving its caches with the success of naive kangaroo
rats searching for these same caches. Nine wild-caught kangaroo rats were allowed to search for caches
that were distributed in the same spatial pattern as that created by one kangaroo rat from the first
experiment. The naive subjects found significantly fewer caches than had the cache owner in the same
length of time. This suggests that the use of spatial memory by a Merriam's kangaroo rat to relocate
its food caches gives it a competitive advantage over other kangaroo rats that may be searching for its
caches.
Foragers commonly store surplus food for later
consumption (Vander Wall 1990). Such behaviour
should only be adaptive if animals that store food
are more likely to retrieve the food caches than
are their competitors (Andersson & Krebs 1978).
When an animal scatter hoards, or places small
surface caches in scattered locations (Morris
1962), such caches are usually not defended and
appear vulnerable to competitors. Indeed, artificially created caches disappear rapidly (Stapanian
& Smith 1978; Thompson & Thompson 1980;
Sherry et al. 1982) and such disappearances have
been attributed to cache pilfering, i.e. the removal
of cache contents by individuals other than the
owner.
In the face of competition from pilferers, a foodstorer must somehow ensure that it will retrieve a
sufficient number of its caches. It might do this by
remembering the precise locations of its caches.
The ability to remember cache locations has been
demonstrated in some species of food-storing
birds (marsh tit, Parus pahlstris, Sherry et al.
*Present address: Department of Psychology, University
of Utah, Salt Lake City, UT 84112, U.S.A.
0003-3472/92/040585 + 09 $03.00/0
1981; Clark's nutcracker, Nucifraga cohtmbiana,
Vander Wall 1982) and mammals (red fox, Vulpes
vulpes, Macdonald 1976; grey squirrel, Sciurus
carolinensis, Jacobs & Liman 1991). In the grey
squirrel study, squirrels preferentially retrieved
nuts from their own cache sites rather than from the
cache sites chosen by other squirrels. However,
because each cache site contained a buried nut, it is
not clear whether they could have found the precise
location without the odour cues emanating from
the cache (Jacobs & Liman 1991).
Whether a mammalian food-storer can find a
cache location without the use of odour cues is not
known. The present study addressed this question
in Merriam's kangaroo rats, Dipodomys merriami.
Merriam's kangaroo rat is a small (35 g) granivorous desert rodent that scatter hoards seeds over
its home range (Daly, Jacobs & Wilson, in press); it
is not known whether it can remember the locations
of these caches. In experiment 1, kangaroo rats
were allowed to search for caches they had made,
only half of which contained seeds. If a kangaroo
rat can remember the locations of its caches, it
should be more likely to search sites where it had
cached seeds than sites where it had not cached
9 1992 The Association for the Study of Animal Behaviour
585
Anhnal Behaviour, 43, 4
586
secds. Moreover, if a kangaroo rat does not need
odour cues to find its caches, it should be equally
likely to search cache sites containing seeds and
sites that did not contain seeds.
A second experiment addressed the question of
whether a kangaroo rat's knowledge of its cache
locations gives it a competitive advantage over
other kangaroo rats searching for its caches. This
advantage has been predicted (Andersson & Krebs
1978), but not empirically demonstrated.
GENERAL METHODS
Subjects
Adult kangaroo rats were live-trapped in the
vicinity of Palm Desert, California and Tucson,
Arizona, and transported to the laboratory 1-9
months before the experiment began; I assumed
that such subjects would be experienced cachers,
retrievers and pilferers. Kangaroo rats were housed
singly in plastic cages measuring 48 x 27 x 20 cm,
filled with a 3-em layer of sand for bedding and
dustbathing and containing a 15 x 9 x 6cm brown
glass jar as a nest chamber. They were fed a mixture of grains (millet, wheat and rolled oats) and
fresh spinach, and had continual access to water.
They were kept under a partially reversed 12:12
light:dark cycle, the lights being extinguished at
1200 hours. Because kangaroo rats are nocturnal,
this increased their activity during a portion of
normal daylight hours. Trials were run between
1200 and 2200 hours.
Apparatus
The test arena was a rectangular wooden
enclosure (2 x 1 x 0.6 m), the floor of which was
composed of eight galvanized steel plates (0.5 x
0.5 m each). Four of the eight plates were perforated with 16 holes, measuring 2-5 cm in diameter.
A small plastic cup was placed into each hole for
caching; these were termed the 'caching plates'. Six
black plastic tunnels (plastic pipe, 5 cm in diameter,
cut in half or plastic pipe elbows, 5 cm in diameter)
were placed at regular intervals around the arena,
against the wall and in the corners, to serve as
hiding places. The cups were filled with white sand
and capped with blue poker chips and served as
discrete artificial cache sites, a design adapted from
Kamil & Balda (1985). The caching cups will hereafter be referred to as 'cache sites'. A large number
ofcache sites were used to increase the difficulty of
orienting correctly to the cache sites during cache
retrieval. The poker chips increased the conspicuousness of any caching or retrieval behaviour and
facilitated the analysis of videotaped records. The
arena was situated in a small room illuminated by
low-intensity red lights which shed sufficient illumination for video-recording. Visual cues outside of
the arena included the light source hanging above
the arena, the video camera mounted on one wall
and shelves on the opposite wall. Auditory cues by
which the kangaroo rats could have oriented themselves, such as conversations occurring outside the
experimental room were also available.
I cleaned the arena before each trial and between
the caching and retrieval phase of each trial to
remove any odours, such as scent marks near
caches, by which a kangaroo rat could relocate
caches without remembering cache locations. First,
I washed the walls with a strong detergent, and then
either replaced or turned over the steel plates which
made up the floor. I sifted the sand used during the
caching phase, mixed it in a large pan, replaced the
sand into new plastic cups, and covered each cup
with a washed and dried poker chip.
EXPERIMENT1
Individual kangaroo rats were allowed to cache
shelled sunflower seeds in the arena. After 24 h, the
subject was allowed to search for its cache sites,
only halfofwhich contained seeds. If kangaroo rats
remembered caches according to their location,
then they should be equally likely to search filled or
emptied cache sites, and they should only search
sites where they had cached. A fourth category
of site caches not made by the subject, was not
included in this study. Although my hypothesis
would predict that kangaroo rats would not search
such sites, subjects would have been encouraged to
search locations where they had not cached, which
would have obscured estimates of their accuracy in
returning only to places where they had cached.
Each subject received six trials; a trial consisted
of a caching phase followed by a retrieval phase.
Eight kangaroo rats (four males and four females)
cached seeds regularly in the sand-filled cups, and
data from these subjects were used to estimate the
accuracy of their memory for specific locations. An
additional six kangaroo rats (five males and one
female) cached only on the surface ofthe arena, not
Jacobs: Kangaroo rat memory
in the cups, and data from their trials were used to
examine their choice of substrates searched during
retrieval.
Methods
Caching phase
Each subject was partially food-deprived for
24 h, during which period its intake was reduced
from an ad libitum supply o f its normal diet to a
smaller amount (2 g) of a less preferred food (rolled
oats). I then released the kangaroo rat into the
arena to eat and cache seeds from a dish of 100
shelled sunflower seeds, situated in the centre of the
arena. I removed the subject from the arena after all
the seeds had been taken, or after 2 h had elapsed,
whichever came first. Subjects were then partially
food deprived until the retrieval phase, 24 h later. I
did not wash the plastic tunnels and the food dish,
and in fact handled them minimally to avoid disturbing scent marks, replacing them in exactly the
same positions as they had occupied during the
caching phase. These tunnels thus served as a set
of familiar olfactory and visual landmarks in the
arena, so that the kangaroo rat would be more
likely to recognize the arena as the same place it
had cached the day before, despite the cleaning
procedures.
To reduce the repeated use of the same sites, I
systematically changed the configuration of available cache sites for each of six trials (Fig. 1). One
subject (male 613) was mistakenly given the third
configuration twice (third and sixth trials). However, despite experiencing the same environment
twice, he cached in non-overlapping distributions
during these two trials, and thus his earlier
memories of the third trial could not have improved
his retrieval accuracy in the sixth trial. The intertrial interval was 10-2+0"7 days (range=4-25),
excluding one long inter-trial interval for each of
two subjects (132 and 178 days).
Retrievalphase
I replaced half of the caches (or the majority, in
the case of an odd number of caches) in the cache
sites for the retrieval. As satiation would reduce the
motivation to search for food, only two seeds were
replaced in each of the chosen cache sites. These
seeds were taken from those originally cached at
that site so that the kangaroo rat, encountering
seeds that it had previously placed at that site,
587
would be more likely to recognize the cache as its
own. This was important because kangaroo rats
became reluctant to cache if they did not find any of
their own caches during retrieval (personal observation). I placed the seeds at the bottom of the new
cup, and covered them with sand and a clean poker
chip.
I then released the subject into the arena for
retrieval of its caches, only half of which actually
contained seeds. Its retrieval behaviour was videotaped for 30 min, and then it was allowed to search
the arena for an additional 30 min, to ensure it had
had time to check each of its former cache sites. I
used the videotaped record of the retrieval phase
to calculate the percentage of each type of site
(emptied cache site, refilled cache site and unused
cache site) searched during the first 15min of
retrieval, although most caches were retrieved in
the first few minutes of access to the arena. I defined
a search as the first displacement of the poker chip
by the kangaroo rat's head.
Results and Discussion
Kangaroo rats cached seeds in the sand-filled
cups in 4.9 _+0"3 (.Y+_SE) of six trials, yielding a total
of 39 trials by eight subjects ( N = 8 for all results). In
three of these trials, kangaroo rats placed only one
cache in a cup. Because this disallowed partial
replacement o f caches, these trials were excluded
from the analyses of retrieval accuracy. In an
additional four trials, kangaroo rats only cached
seeds on the surface of the steel plates, usually along
the arena walls. Regardless of where seeds were
cached, the subject was always allowed to retrieve
its caches the next day. In the case of surface
caches, some seeds were replaced for the retrieval
phase, to encourage the subject to continue caching. Kangaroo rats cached in 5-4_+ I-0 cups per trial
( r a n g e = l - 1 4 ) , placing 11-5+0.9 seeds in each
cache and 2.8-1-0.3 caches on each caching plate.
Thus, caches were typically distributed in clumps,
each containing several caches, on some but not all
of the four plates available for caching. This is illustrated in Fig. 1 by the cache distributions of three
subjects.
To test the hypothesis that kangaroo rats can
relocate caches without detecting the odour of
buried seeds, the percentage of cups searched was
compared among the three categories of cups: noncache sites, emptied cache sites and refilled cache
sites. As seen in Table I, kangaroo rats searched a
Anh~tal Behaviour, 43, 4
588
I Femo~elOI I
I Mo~e61~I
"!i!i!i!
::':':':
!:i:i:i"
::i:i:i:
iiiiiii.', ii!iiiii
. . . :i:i':,
:':::':,
"""'-"
!:i:i:i:.:.9
::
Mole 105 J
9 . :i:i'i:i
D
9
:""'
:!i!i!:
=.
,,,,,,. ,,,,,
;-;-;-:-;,;-
i
.,,,,.-*,,.,.-,
9
9
,',',',
:,-,-,-,
9
m *!
P
;';';';
,,,,,,, .,,,,,,,
,,,,,,,
" " ",',',"
,~,~,~.,~,~,~.
i!i!i!iE!!!!!!i 9
......
:,:rzr:' .....
?,~,Z',~' 9
~
9
~
.','."
",',',"
9
.,, ,,, ,,,,,,,
"": ,""
;
;,;,;,:
9
*--- ,,-,-,,.
i,;,7,. ~;,;,;,.
r
" '9
,;-i,1,;
. . ,.
,~,-~,-~-,:,-,,,,
Q
O0
Figure I. The distribution of caches produced by three kangaroo rats in experiment 1. The floor of the caching arena
consisted ofeight metal plates, four ofwhich were unperforated (non-caching plates) and four ofwhich were perforated
and held 16 sand-filled cups for caching, evenly spaced in a 4 x 4 matrix. The spatial configuration ofcaching plates ([])
and non-caching plates ([]) was changed between trials, as shown. Cache sites ( 0 ) were defined as cups in which the
kangaroo rats buried seeds.
significantly greater percentage of sites in which
they had cached seeds, than sites in which they
had not cached seeds (two-tailed t-test: t = 11.15,
d f = 1 4 , P<0.0001). They were not more likely
to search cache sites with seeds; there was no significant difference between the percentage of filled
and emptied cache sites that were searched
(two-tailed t-test: t=0.919, d f = 1 4 , P=0.624).
Moreover, they searched a significantly greater percentage of emptied sites than unused sites (twotailed t-test: t=7.21, df= 14, P<0.0001). Because
these sites were physically identical, it appears that
kangaroo rats did not need to use olfaction to find
caches.
R a t h e r t h a n olfaction, m e m o r y for cache
locations appears to play an important role in
cache retrieval. If a kangaroo rat can remember
its cache locations, it should be more likely to
search cache sites than unused sites. Because of the
small number of caches made per trial, most sites
(58.3__+1.0 of 64 sites, or 91.1%) were unused.
However, on average 7 7 . 8 + 4 . 2 % of the first sites
that a kangaroo rat searched (up to the n u m b e r of
caches it had made) were cache sites.
The probability that a subject's retrieval of its
caches was greater than that predicted by cache
availability can be calculated from the probability
of a subject making the observed n u m b e r of
Jacobs: Kangaroo rat memory
Table I. Accuracy of cache retrieval, as estimated by the
mean number and percentage of cache and non-cache
sites searched, relative to the total number of that type of
site available, during the first 15 min of retrieval (N= 8)
Number of
sites searched
Percentage of
sites searched
Type of site
..Y
sE
X
SE
Refilled cache
Emptied cache
Unused
(non-cache)
3"0
2-3
8.2
0.5
0.4
2-0
93-3
86-9
14-1
3.0
6.5
3-5
Percentage of sites is calculated as the number searched
divided by the number available for searching at the
beginning of the trial.
correct choices, given the availability of cache sites
and non-cache sites (Polimeni & Straight 1985; see
Appendix). Because the availability of cache sites
and non-cache sites changes as more sites are
searched, this method sums the changing probabilities of a cache site being chosen, given the availability of caches and non-cache sites at each point
during retrieval. The process continues up to the
maximum number of correct choices, which is the
total number ofcaches available at the start.
This analysis yielded the following result: in 34 of
39 trials in which the subject cached in at least one
cup, kangaroo rats were significantly more likely to
search cache sites than predicted by the availability
of cache sites (P<0-02 for each trial). In four of
the remaining five trials, retrieval of caches was not
significantly different from that predicted by availability ( P = 0.062 for each trial). In these four trials,
each subject made two caches and searched one
cache site and one non-cache site in the first two
searches. In the fifth case, the kangaroo rat showed
no evidence of searching cache sites before noncache sites ( P = 1.0). He had made six caches and all
of his first six choices were non-cache sites (male
105, fifth trial, Fig. 1). This lack of accuracy was
unique; it was the only trial in which such a pattern
of retrieval occurred. The poor performance of this
subject was especially striking, as in his four previous trials he had consistently retrieved all of his
caches before investigating any non-cache sites.
During this trial, this subject also searched a
greater number of non-cache sites than had any
other subject during any other trial (42, or 72"4% of
those available, compared to the group mean of
589
14.1%) and he searched 20 of these non-cache sites
before he found all o f his six caches. It may be
significant that the locations of his first six choices
resembled a mirror image o f his cache sites. This
suggests that he might have remembered the spatial
configuration of the caches but somehow became
disoriented and hence searched the wrong end of
the arena.
As mentioned earlier, in some trials the subjects
did not cache in the cups but instead cached on the
surface of the arena floor, and these trials were
excluded from the previous analysis. These data
were combined with data from cup-cache retrievals
to test the following hypothesis: if kangaroo rats
remember cache locations, then subjects who had
not cached in the cache cups should not search
cache cups during retrieval. Indeed, in 9 of 12 trials
in which subjects had only cached on the surface,
such subjects d i d not search any cups during
retrieval; in the remaining three trials, subjects
searched only one or two cups. When data from
surface-cache trials and cup-cache trials were combined, there was a significant positive correlation
between the number of cups used for caching
(range=0-14) and the number of unused cups
searched (range=0-42) during retrieval ( N = I 4 ,
r=0-35, P=0-011). Thus, kangaroo rats only
searched cups when they previously had cached
seeds in cups, again suggesting that memory plays
an important role in cache retrieval.
If, in the course of retrieving its caches, a
kangaroo rat searches an unused site before it has
retrieved all of its caches, then this may be interpreted as an error in locating caches. However, ifit
only searches unused sites after it has found all of
its caches, then such a search cannot be errors in
retrieval but must be otherwise motivated, such as
the need to explore. Similarly, if non-cache site
searches represent errors, then one would predict
that kangaroo rats would be more likely to search
sites near caches than sites further away or in
areas where no caches had been made. To test
this hypothesis, searches of non-cache sites were
categorized into those searched before and after all
caches were retrieved, and by their location relative
to the nearest cache (Table II). Searches of sites on
plates that contained caches were defined as 'near
cache' and searches of sites on plates without
caches were defined as 'far from cache'.
Overall, kangaroo rats were more likely to search
non-cache sites after they had retrieved all of
their caches than before (two-tailed t-test: t=2.28,
Anhnal Behaviour, 43, 4
590
Table 1I. Searches of non-cache sites during the first
15 min of retrieval (N= 8)
Location of
non-cache site
searches
(I) All non-cache sites
(2) Plates with caches
(3) Plates without caches
Before
retrievals
.eV
SE
2-3 0.6
1-3 0.2
I-0 0-5
After
retrievals
,~
SE
6.4
4.3
2.1
1.7
l.l
0-7
Searches occurred either (I) before or after the kangaroo
rat had retrieved all of its caches, (2) on plates that contained caches during that trial, or (3) on plates that did not
contain caches.
df= 14, P=0"037). When searches were divided
into those near and far from caches, the same
result was found: subjects searched more sites after
retrieving all caches than before (two-tailed t-test:
t=2.65, df= 14, P=0-018). Despite a trend in the
same direction, there was no significant difference
in the number of 'far from cache' sites searched
before and after all retrievals had been made (twotailed t-test: t = l . 2 1 , df=14, P=0.246). These
results imply that kangaroo rats retrieve all oftheir
caches first and then search other non-cache sites
nearby, particularly those near current cache sites;
sites distant from any cache site may be searched
either before or after all the caches have been
retrieved. Because most searches of non-cache sites
occurred after all retrievals had been made, these
should not have interpreted as inaccurate retrieval
attempts. This pattern of search may be an adaptive
response to the heavy loss of caches induced by the
experimental design (i.e. I removed half of the caches
completely and reduced the remaining caches to two
seeds each). Therefore, the kangaroo rats may
simply have been rechecking the general area in
which they had previously cached seeds.
EXPERIMENT 2
In this experiment, kangaroo rats were allowed to
find and eat seeds cached by another kangaroo rat
(the cache owner), and their foraging efficiency was
then compared with the retrieval efficiency of the
cache owner, as previously determined in experiment 1. I predicted that the cache owner, remembering the locations of its caches, would have found
more caches in the same length of time than would a
non-owner.
Methods
Subjects
Thirteen kangaroo rats (seven males and six
females) served as non-owners. Five of these had
completed the first experiment and the other eight,
trapped at the same time and maintained under
the same conditions, had only cached seeds a few
times in the arena. There was no difference in the
results between these two groups and their data
were subsequently combined.
Procedure
After a 24-h period of limited food deprivation
(up to 2 g rolled oats), each subject was allowed
15 rain to find seeds buried in one of five cache distributions, on five consecutive days. Non-owners
were thus given five opportunities to find the caches
of one individual (male 613). The caches were
placed in arrays of the first five distributions that
this individual had produced during experiment 1,
and were presented in the same order as shown in
Fig. 1.
Before each trial the arena was cleaned in the
manner described for experiment 1, and two shelled
sunflower seeds were placed at the bottom of a
sand-filled cup in the location of each cache made
by male 613. The non-owner kangaroo rat was then
released into the arena and allowed to search for
and eat or remove seeds. After 15 min, the subject
was removed, gently relieved of any seeds in its
cheek pouches, and returned to its holding cage.
The number of seeds it had eaten was calculated
from the number of seeds still cached in the arena,
and its ration of roiled oats for the intervening
period before the next day's trial was adjusted
accordingly. The total number of caches available
ranged from six to nine; thus 12-18 seeds were
available in the arena during each trial. This was
usually in excess of what a kangaroo rat would
eat; in some cases, kangaroo rats would re-bury
excavated seeds at other sites.
Each trial was recorded on videotape and all cup
searches, defined as the first displacement of the
poker chip by the kangaroo rat's head, were scored
to calculate the number and percentage of cache
sites and non-cache sites that were searched.
Results and Discussion
Because the percentage ofcaches found in 15 min
did not differ significantly between the inexperienced and experienced kangaroo rats ( M a n n -
Jacobs: Kangaroo rat memory
Table IlL The mean percentage of non-cache sites
searched and cache sites found by non-owner kangaroo
rats, one individual cache owner and all cache owners in
experiment !
Cache s i t e
Subject
N
,~
SE
Non-cache site
,~
591
t = 5-88, df= 13, P < 0-0001). Thus, when searching
the same number of sites, the average non-owner
found significantlyfewer caches than did the owner
of the caches, and chose significantly fewer cache
sites in the beginning of their searches than did the
average cache owner.
SE
GENERAL DISCUSSION
Non-owners
Cache owner
(male 613)
All owners
9
i
68.7
100.0
10.3
0
13.6
14.2
3.2
3-1
8
92.6
3'2
14.1
3-5
Whitney U-tests: U = 17.0, P=0.328), these data
were combined. Also, because 4 of the 13 subjects
searched fewer than five cups during the five trials
their search behaviour could not be interpreted and
the data from these subjects were excluded.
Non-owner kangaroo rats found, on average,
3"8___0"8 cache sites (Table III). This was 31.3%
fewer cache sites containing seeds than the cache
owner had found in experiment 1, a significant
difference (one-sample t-test: t=3.05, df=8,
P<0.015). The non-owners' success in finding
caches can also be measured by restricting the
analysis to the number ofcorrect choices made, i.e.
cache sites chosen up to the number ofcaches available. For example, seven caches were available in
the arena on the first trial, and in her first seven
searches, non-owner female 103 searched three
cache sites and four non-cache sites, or 43% correct. Only those non-owners that had searched at
least as many sites as there were caches available
could be used in this analysis, reducing the sample
from nine subjects to seven. The average scores of
the non-owners were compared with the average
score of the owner and with the average scores of
all cache owners from experiment 1. This latter
comparison gives a more conservative estimate of
an owner's advantage relative to a cache owner, as
the cache owner in this experiment (male 613) had
displayed a higher than average retrieval accuracy
in experiment 1.
Non-owners had significantly lower success
searching for caches than the cache owner. Nonowners made 39.4% fewer correct choices than
had the cache owner, a significant difference (one
sample t-test: t=23.5, df=6, P<0.0001). Even
when compared to the average owner in experiment
I, non-owners made 28-0% fewer correct choices,
also a significant difference (two-tailed t-test:
This study addresses the question of whether
kangaroo rats can remember the spatial locations
of their food caches. The first experiment demonstrates that kangaroo rats retrieve their caches
based on the spatial locations of the caches and not
by the odour of buried seeds, as they searched equal
numbers of emptied and refilled cache sites during
retrieval. Because they searched a lower percentage
of unused sites than emptied caches, and because
these sites differed only by their location, the
kangaroo rats must have distinguished them, and
hence remembered them, by their location. Furthermore, non-cache sites were usually searched after
all cache sites had been searched. Thus, the
search of non-cache sites might be interpreted as
exploration or as additional attempts to find pilfered caches, and not as an inaccurate orientation
to remembered cache locations.
The results of the second experiment indicate
that detecting a cache by its odour, though effective, is not as efficient as remembering its location.
The cache owner found 31% more caches than did
the average non-owner, and searched fewer sites.
Thus, the owner acquired more food for less effort
than a non-owner. The owner strategy is not necessarily more efficient in total energy expenditure,
however, because the owner had already invested
energy by caching the seeds and by remembering
their locations. The extra seeds discovered by
the owner simply might have offset its previous
expenditure of effort. On the other hand, the value
of a cache is not only determined by its energetic
content and the energy expended in caching and
retrieval, but also by the availability of alternative
food sources at the time of retrieval. Caches may
increase in value as other foods become more
scarce or when foraging is costly, such as during
inclement weather or periods of high predation risk
or intense competition from other foragers. Under
these conditions, the cache owner strategy could be
highly profitable.
This study presents the first evidence that memory
for cache locations has immediate adaptive value for
Anhnal Behaviour, 43, 4
592
a scatter hoarder such as the Merriam's kangaroo
rat. The probability of pilfering may be particularly
high for this species. The caches of Merriam's
kangaroo rats are routinely pilfered by neighbouring kangaroo rats and other species ofgranivorous
rodents (Daly, Jacobs & Wilson, in press). Competition for food between desert rodents may be
a major determinant of the structure of desert
rodent communities (Price & Brown 1983). A
kangaroo rat's memory for its cache locations may
serve several functions: if it can remember
individual sites, then it can distribute caches in
arrays that are unpredictable to its competitors, or
change the location and density of its caches in
response to pilfering. This latter response has
been demonstrated in a scatter hoarding bird, the
magpie, Pica pica, which spaces its caches more
widely in higher risk habitats, i.e. where caches are
pilfered more quickly (Clarkson et al. 1986). Other
studies have also shown that the density of caches
is a key factor in determining the rate of their
disappearance (Stapanian & Smith 1978; Sherry et
al. 1982).
A second ecological factor that may select for
good spatial memory in Merriam's kangaroo rats is
the reduced risk of predation, especially while
retrieving caches. In the second experiment, owners
retrieved more caches than non-owners in the same
length of time. The owner's advantage reported
here may be a minimum estimate, because the
arena was small and easily searched, and the
seeds were larger, and perhaps more pungent,
than those cached in the desert (Brown et al.
1979). Under natural conditions, where kangaroo
rats range widely (Behrends et al. 1986) and scatter much smaller caches over large areas (Daly,
Jacobs & Wilson, in press), owners who remember cache locations may significantly reduce
their chance of predation while retrieving caches.
Kangaroo rats forage, and perhaps retrieve
caches, in open sites where they may be more vulnerable to predation than other granivorous
rodents, such as pocket mice (e.g. Perognathus),
which forage near and under shrubs (Brown &
Lieberman 1973; Lemen & Rosenzweig 1978;
Thompson 1982). If a kangaroo rat's memory of
its cache locations allows it to retrieve caches
more quickly, this should reduce its risk of predation, relative to the risk faced by a pilferer.
Thus, kangaroo rats may have more than one
reason to remember cache locations: to retrieve
food from scattered locations before the cache is
found by a pilferer or before the cacher is found
by a predator.
APPENDIX
The following analysis is taken from Polimeni & Straight (1985). Let c = t h e number of caches and let
r = the number of correct retrievals in the first c searches. The total number of available sites is 64, and the
probability of making one correct retrieval from c caches is:
[c!/r!(c- r)!] x {(64 - c)!/(c- 0![(64 - c) - ( c - r)]! }
[64!1d(64-c)!]
The probability of retrieving the observed number of caches is the sum of the above probability and the
similarly computed probability for more caches retrieved, r + 1, r + 2 . . . c, which is the maximum number
ofcorrect choices. This technique yields the values reported in the text.
ACKNOWLEDGMENTS
I thank D. Sherry and M. Daly for advice,
encouragement and comments on the manuscript;
W. Spencer, M. Daly and M. Wilson for help in
trapping and shipping kangaroo rats; L. Zanette for
technical assistance; S. Gaulin and M. Fitzsimmons
for statistical advice; L.-A. Giraldeau, J. Hogan
and S. Shettleworth for helpful discussions; D.
Dearing, R. Herz, S. Jenkins, G. Michener, S.
Mizumori and J. Murie for comments on the manu-
script. This research was supported by a N A T O
Postdoctoral Fellowship and N I M H Postdoctoral
Fellowship 1 F32 MH09701-01 to the author,
and Canadian Natural Sciences and Engineering
Research Council grants to D. Sherry and M. Daly.
REFERENCES
Andersson, M. & Krebs, J. R. 1978. On the evolution of
hoarding behaviour. Anita. Behav., 26, 707-71 I.
Jacobs: Kangaroo rat memory
Behrends, P., Daly, M. & Wilson, M. I. 1986. Range
use patterns and spatial relationships of Merriam's
kangaroo rats (Dipodomys merriamO. Behaviour, 96,
187-209.
Brown, J. H. & Lieberman, G. A. 1973. Resource utilization and coexistence of seed-eating desert rodents in
sand-dune habitats. Ecology, 54, 788-797.
Brown, J. H., Reichman, O. J. & Davidson, D. W. 1979.
Granivory in desert ecosystems. A. Rev. EcoL Syst., 10,
201-227.
Clarkson, K., Eden, S. F., Sutherland, W. J. & Houston,
A. I. 1986. Density dependence and magpie food
hoarding. J. Anbn. EcoL, 55, 111-121.
Daly, M., Jacobs, L. F. & Wilson, M. I. In press. Scatter
hoarding by kangaroo rats (Dipodomys merriamO and
pilferage from their caches. Behav. EcoL
Jacobs, L. F. & Liman, E. R. 1991. Grey squirrels
remember the locations of buried nuts. Anita. Behav.,
41,103-110.
Kamil, A. C. & Balda, R. C. 1985. Cache recovery and
spatial memory in Clark's nutcrackers (Nucifraga
columbiana). J. exp. psychoL Anbn. Behav. Proc., II,
95-111.
Lemen, C. A. & Rosenzweig, M. L. 1978. Microhabitat
selection in two species of heteromyid rodents.
Oecologia ( BerL), 33, 127-135.
Macdonald, D. W. 1976. Food caching by red foxes and
some other carnivores. Z. TierpsychoL,42, 170-185.
593
Morris, D. 1962. The behaviour of the green acouchi
(Myoprocta prattO with special reference to scatter
hoarding. Proc. ZooL Soc. Lond., 139, 701-732.
Polimeni, A. D. & Straight, H. J. 1985. Foundations of
Discrete Mathematics. Monterey, California: Brooks/
Cole.
Price, M. V. & Brown, J. H. 1983. Patterns of morphology
and resource use in North American desert rodent
communities. Great Basin Nat. Mem., 7, 117-134.
Sherry, D. F., Avery, M. & Stevens, A. 1982. The spacing of
stored food by marsh tits. Z. TierpsychoL,58, 153-162.
Sherry, D. F., Krebs, J. R. & Cowie, R. J. 1981. Memory
for the location of stored food in marsh tits. Anbn.
Behav., 29, 1260-1266.
Stapanian, M. A. & Smith, C. C. 1978. A model for seed
scatterhoarding: coevolution of fox squirrels and black
walnuts. Ecology, 59, 884-896.
Thompson, D. C. & Thompson, P. S. 1980. Food habits
and caching behavior of urban grey squirrels. Can. J.
ZooL, 58, 701-710.
Thompson, S. D. 1982. Microhabitat utilization and
foraging behavior of bipedal and quadrupedal
heteromyid rodents. Ecology, 63, 1303-1312.
Vander Wall, S. B. 1982. An experimental analysis of
cache recovery in Clark's nutcracker. Anita. Behav., 309
84-94.
Vander Wall, S. B. 1990. Food Hoarding in Animals.
Chicago: University of Chicago Press.